Anthropic Claude integrationIn development
Put Claude to work on your production issues
The Anthropic Claude integration is in development. It is designed so analyses run on your own Anthropic API key, with ROOKON packaging the evidence (traces, session events, relevant source files) into a focused request for Claude.
The problem
Good analysis needs the right context, not all of it
A useful diagnosis often needs several source files next to the stack trace and the steps a user took. Pasting that into a chat window by hand is slow and easy to get wrong.
- Copying traces and files into a chat by hand
- Missing the user's steps that triggered the bug
- Inconsistent answers from ad-hoc prompts
- No record of what was analyzed or why
Live demo 1 of 3
Connect your Anthropic key safely
Watch envelope encryption happen to a sample Anthropic key. The plaintext leaves memory as soon as it’s encrypted. Runs in your browser and sends nothing.
- Use the sample key
- Revoke it afterwards
rookon · settings · AI provider · Anthropic Claude
Simulated dataConnect Anthropic ClaudeIn development
Don’t paste a real key. This demo runs only in your browser and sends nothing. Use the sample key instead.
Envelope encryption
When you save a key:
- a fresh data key encrypts it (AES-256-GCM),
- a master key that never touches the database wraps the data key,
- only ciphertext and a masked hint are stored,
- the worker decrypts it for a single analysis request, then forgets it.
The API returns only maskedKey after saving, never the key itself, to anyone.
How ROOKON helps
What you get
Your Anthropic account
Create a key in a dedicated Anthropic Console workspace so ROOKON's usage and spend limits are kept separate from your other work.
Curated context
ROOKON selects the trace, breadcrumbs, session summary and the source files on the failing path, and leaves out everything else.
Provider per project
Use Claude for one project and another provider for a different one. The workflow and output format stay identical.
Rate-limit aware
Analyses run as background jobs and are retried with backoff if the provider is busy or rate-limited, so nothing in your app waits on them.
Encrypted and never exposed
Keys are envelope-encrypted at rest, decrypted only in the worker for the duration of a request, and never returned by the API once saved.
Live demo 2 of 3
Pick a Claude model per task
After connecting, choose which of your models diagnoses issues and which writes the code. Model names are examples: ROOKON lists the models your own key can use.
- Change the code generation model
- Set review revisions back to “Same as fallback”
rookon · workspace settings · AI models
Simulated dataConnected providers
Model names below are examples. In ROOKON the list is fetched from your provider account with your key.
| Task | Model | Est. per run |
|---|---|---|
Diagnosis A fast, inexpensive model is usually enough. | ~$0.022 | |
Code generation Pick your strongest coding model. This is where quality matters most. | ~$0.87 | |
Review revisions Defaults to your code generation model. | Uses Code generation: your-strongest-coding-model | ~$0.36 |
Marketing copy Defaults to your diagnosis model. | Uses Diagnosis: your-fast-model | ~$0.006 |
Support replies Defaults to your diagnosis model. | Uses Diagnosis: your-fast-model | ~$0.004 |
Content writing Defaults to your marketing copy model. | Uses Diagnosis: your-fast-model | ~$0.007 |
The model you pick for code generation writes every proposed change. Model routing is in development with the rest of ROOKON.
What happens during a Claude analysis
- 01
Collect
ROOKON gathers the evidence attached to the issue.
- 02
Trim
Sensitive values are filtered and only relevant files are included.
- 03
Ask
A structured request is sent to Claude using your key and chosen model.
- 04
Present
The recommendation and any proposed diff appear on the issue for review.
Live demo 3 of 3
Cap what Claude costs you
Set a monthly budget for analyses on your Claude key and decide what runs automatically. Prices from the model picker above feed the estimate. Simulated.
rookon · workspace settings · AI budget
Simulated dataBilled by your AI provider, on your key. ROOKON stops before the cap.
Your model’s price per 1M tokens
Example prices. Check your provider’s current pricing for your model.
Simulated day 1 of 30
Live$0.00 of $25 this month
on trackprojected month-end $0.000 tokens0 automatic · 2 waiting
Detected issues (newest first)
- infoRarely used setting causes console warningday 1This issue is below the severity your autopilot mode analyzes automatically.
- infoImage without dimensions on /blogday 1This issue is below the severity your autopilot mode analyzes automatically.
Simulated month. Most impactful issues are analyzed first; ROOKON reserves the worst case before each call so a single analysis can’t push you over the cap.
Try: Pick a pricier diagnosis model above · Raise the cap
FAQ
Common questions
Is the Claude integration available?
It is in development, and this page will say so as soon as you can connect an Anthropic key.
Which Claude models will be supported?
You will be able to choose from the models available to your API key. The supported list will be published in the documentation at launch.
How does Anthropic handle the data ROOKON sends?
Requests are made under your own Anthropic account, so your agreement with Anthropic governs them. Review Anthropic's commercial terms and data policies for how API inputs are handled.
Can I cap how much ROOKON spends?
Yes, in two places: set spend limits on the Anthropic Console workspace that holds the key, and choose in ROOKON which issues are analyzed automatically and which wait for a manual request.
Give Claude the context it needs
Sign up and connect your Anthropic key in the setup journey.